Displaying 3 results from an estimated 3 matches for "coverage_mt_vec".
2014 Apr 25
2
[LLVMdev] multithreaded performance disaster with -fprofile-instr-generate (contention on profile counters)
...to the source code is simply not going to work with
>> -fprofile-instr-generate. The mapping from counter values to the
>> user-visible execution counts is complex and relies on detailed knowledge of
>> the clang ASTs.
>>
>>
>>
>> % clang++ -O2 -lpthread coverage_mt_vec.cc && time ./a.out
>> TIME: real: 0.219; user: 0.430; system: 0.000
>> % clang++ -O2 -lpthread -fprofile-instr-generate coverage_mt_vec.cc && time
>> ./a.out
>> TIME: real: 3.743; user: 7.280; system: 0.000
>>
>> % cat coverage_mt_vec.cc
>>...
2014 Apr 23
4
[LLVMdev] multithreaded performance disaster with -fprofile-instr-generate (contention on profile counters)
...kinds of instrumentation.
Using the PC to map to the source code is simply not going to work with -fprofile-instr-generate. The mapping from counter values to the user-visible execution counts is complex and relies on detailed knowledge of the clang ASTs.
>
>
> % clang++ -O2 -lpthread coverage_mt_vec.cc && time ./a.out
> TIME: real: 0.219; user: 0.430; system: 0.000
> % clang++ -O2 -lpthread -fprofile-instr-generate coverage_mt_vec.cc && time ./a.out
> TIME: real: 3.743; user: 7.280; system: 0.000
>
> % cat coverage_mt_vec.cc
> #include <pthread.h>
>...
2014 Apr 18
4
[LLVMdev] multithreaded performance disaster with -fprofile-instr-generate (contention on profile counters)
On Fri, Apr 18, 2014 at 12:13 AM, Dmitry Vyukov <dvyukov at google.com> wrote:
> Hi,
>
> This is long thread, so I will combine several comments into single email.
>
>
> >> - 8-bit per-thread counters, dumping into central counters on overflow.
> >The overflow will happen very quickly with 8bit counter.
>
> Yes, but it reduces contention by 256x (a thread